diff --git a/CMakeLists.txt b/CMakeLists.txt
index 3c5856fc59097..35baea9417564 100644
--- a/CMakeLists.txt
+++ b/CMakeLists.txt
@@ -168,7 +168,6 @@ if(NVCC_THREADS AND VLLM_GPU_LANG STREQUAL "CUDA")
   list(APPEND VLLM_GPU_FLAGS "--threads=${NVCC_THREADS}")
 endif()
 
-
 #
 # Use FetchContent for C++ dependencies that are compiled as part of vLLM's build process.
 # setup.py will override FETCHCONTENT_BASE_DIR to play nicely with sccache.
@@ -308,148 +307,70 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
   # Keep building Marlin for 9.0 as there are some group sizes and shapes that
   # are not supported by Machete yet.
   # 9.0 for latest bf16 atomicAdd PTX
-  cuda_archs_loose_intersection(MARLIN_ARCHS "8.0;9.0+PTX" "${CUDA_ARCHS}")
-  if (MARLIN_ARCHS)
+  # Marlin kernels: generate and build for supported architectures
+  cuda_archs_loose_intersection(MARLIN_ARCHS "8.0+PTX" "${CUDA_ARCHS}")
+  optional_cuda_sources(
+    NAME Marlin
+    ARCHS "8.0;9.0+PTX"
+    GEN_SCRIPT "${CMAKE_CURRENT_SOURCE_DIR}/csrc/quantization/gptq_marlin/generate_kernels.py"
+    GEN_GLOB "csrc/quantization/gptq_marlin/kernel_*.cu"
+    SRCS
+      "csrc/quantization/marlin/dense/marlin_cuda_kernel.cu"
+      "csrc/quantization/marlin/sparse/marlin_24_cuda_kernel.cu"
+      "csrc/quantization/marlin/qqq/marlin_qqq_gemm_kernel.cu"
+      "csrc/quantization/gptq_marlin/gptq_marlin.cu"
+      "csrc/quantization/gptq_marlin/gptq_marlin_repack.cu"
+      "csrc/quantization/gptq_marlin/awq_marlin_repack.cu"
+    NO_ARCH_MSG
+      "Not building Marlin kernels as no compatible archs found in CUDA target architectures"
+  )
 
-    #
-    # For the Marlin kernels we automatically generate sources for various
-    # preselected input type pairs and schedules.
-    # Generate sources:
-    set(MARLIN_GEN_SCRIPT
-      ${CMAKE_CURRENT_SOURCE_DIR}/csrc/quantization/gptq_marlin/generate_kernels.py)
-    file(MD5 ${MARLIN_GEN_SCRIPT} MARLIN_GEN_SCRIPT_HASH)
-
-    message(STATUS "Marlin generation script hash: ${MARLIN_GEN_SCRIPT_HASH}")
-    message(STATUS "Last run Marlin generate script hash: $CACHE{MARLIN_GEN_SCRIPT_HASH}")
-
-    if (NOT DEFINED CACHE{MARLIN_GEN_SCRIPT_HASH}
-        OR NOT $CACHE{MARLIN_GEN_SCRIPT_HASH} STREQUAL ${MARLIN_GEN_SCRIPT_HASH})
-      execute_process(
-        COMMAND ${CMAKE_COMMAND} -E env
-        PYTHONPATH=$PYTHONPATH
-          ${Python_EXECUTABLE} ${MARLIN_GEN_SCRIPT}
-        RESULT_VARIABLE marlin_generation_result
-        OUTPUT_VARIABLE marlin_generation_result
-        OUTPUT_FILE ${CMAKE_CURRENT_BINARY_DIR}/marlin_generation.log
-        ERROR_FILE ${CMAKE_CURRENT_BINARY_DIR}/marlin_generation.log
-      )
-
-      if (NOT marlin_generation_result EQUAL 0)
-        message(FATAL_ERROR "Marlin generation failed."
-                            " Result: \"${marlin_generation_result}\""
-                            "\nCheck the log for details: "
-                            "${CMAKE_CURRENT_BINARY_DIR}/marlin_generation.log")
-      else()
-        set(MARLIN_GEN_SCRIPT_HASH ${MARLIN_GEN_SCRIPT_HASH}
-            CACHE STRING "Last run Marlin generate script hash" FORCE)
-        message(STATUS "Marlin generation completed successfully.")
-      endif()
-    else()
-      message(STATUS "Marlin generation script has not changed, skipping generation.")
-    endif()
-
-    file(GLOB MARLIN_TEMPLATE_KERNEL_SRC "csrc/quantization/gptq_marlin/kernel_*.cu")
-    set_gencode_flags_for_srcs(
-      SRCS "${MARLIN_TEMPLATE_KERNEL_SRC}"
-      CUDA_ARCHS "${MARLIN_ARCHS}")
-
-    list(APPEND VLLM_EXT_SRC ${MARLIN_TEMPLATE_KERNEL_SRC})
-
-    set(MARLIN_SRCS
-       "csrc/quantization/marlin/dense/marlin_cuda_kernel.cu"
-       "csrc/quantization/marlin/sparse/marlin_24_cuda_kernel.cu"
-       "csrc/quantization/marlin/qqq/marlin_qqq_gemm_kernel.cu"
-       "csrc/quantization/gptq_marlin/gptq_marlin.cu"
-       "csrc/quantization/gptq_marlin/gptq_marlin_repack.cu"
-       "csrc/quantization/gptq_marlin/awq_marlin_repack.cu")
-    set_gencode_flags_for_srcs(
-      SRCS "${MARLIN_SRCS}"
-      CUDA_ARCHS "${MARLIN_ARCHS}")
-    list(APPEND VLLM_EXT_SRC "${MARLIN_SRCS}")
-    message(STATUS "Building Marlin kernels for archs: ${MARLIN_ARCHS}")
-  else()
-    message(STATUS "Not building Marlin kernels as no compatible archs found"
-                   " in CUDA target architectures")
-  endif()
-
-  # Only build AllSpark kernels if we are building for at least some compatible archs.
-  cuda_archs_loose_intersection(ALLSPARK_ARCHS "8.0;8.6;8.7;8.9" "${CUDA_ARCHS}")
-  if (ALLSPARK_ARCHS)
-    set(ALLSPARK_SRCS
-       "csrc/quantization/gptq_allspark/allspark_repack.cu"
-       "csrc/quantization/gptq_allspark/allspark_qgemm_w8a16.cu")
-    set_gencode_flags_for_srcs(
-      SRCS "${ALLSPARK_SRCS}"
-      CUDA_ARCHS "${ALLSPARK_ARCHS}")
-    list(APPEND VLLM_EXT_SRC "${ALLSPARK_SRCS}")
-    message(STATUS "Building AllSpark kernels for archs: ${ALLSPARK_ARCHS}")
-  else()
-    message(STATUS "Not building AllSpark kernels as no compatible archs found"
-                   " in CUDA target architectures")
-  endif()
+  # AllSpark kernels
+  optional_cuda_sources(
+    NAME AllSpark
+    ARCHS "8.0;8.6;8.7;8.9"
+    SRCS
+      "csrc/quantization/gptq_allspark/allspark_repack.cu"
+      "csrc/quantization/gptq_allspark/allspark_qgemm_w8a16.cu"
+    NO_ARCH_MSG "Not building AllSpark kernels as no compatible archs found in CUDA target architectures"
+  )
 
 
   set(SCALED_MM_3X_ARCHS)
-  # The cutlass_scaled_mm kernels for Hopper (c3x, i.e. CUTLASS 3.x) require
-  # CUDA 12.0 or later
-  cuda_archs_loose_intersection(SCALED_MM_ARCHS "9.0a;" "${CUDA_ARCHS}")
-  if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER 12.0 AND SCALED_MM_ARCHS)
-    set(SRCS
-       "csrc/quantization/cutlass_w8a8/scaled_mm_c3x_sm90.cu"
-       "csrc/quantization/cutlass_w8a8/c3x/scaled_mm_sm90_fp8.cu"
-       "csrc/quantization/cutlass_w8a8/c3x/scaled_mm_sm90_int8.cu"
-       "csrc/quantization/cutlass_w8a8/c3x/scaled_mm_azp_sm90_int8.cu"
-       "csrc/quantization/cutlass_w8a8/c3x/scaled_mm_blockwise_sm90_fp8.cu")
-    set_gencode_flags_for_srcs(
-      SRCS "${SRCS}"
-      CUDA_ARCHS "${SCALED_MM_ARCHS}")
-    list(APPEND VLLM_EXT_SRC "${SRCS}")
-    list(APPEND VLLM_GPU_FLAGS "-DENABLE_SCALED_MM_SM90=1")
-    # Let scaled_mm_c2x know it doesn't need to build these arches
-    list(APPEND SCALED_MM_3X_ARCHS "${SCALED_MM_ARCHS}")
-    message(STATUS "Building scaled_mm_c3x_sm90 for archs: ${SCALED_MM_ARCHS}")
-  else()
-    if (NOT ${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER 12.0 AND SCALED_MM_ARCHS)
-      message(STATUS "Not building scaled_mm_c3x_sm90 as CUDA Compiler version is "
-                     "not >= 12.0, we recommend upgrading to CUDA 12.0 or "
-                     "later if you intend on running FP8 quantized models on "
-                     "Hopper.")
-    else()
-      message(STATUS "Not building scaled_mm_c3x_sm90 as no compatible archs found "
-                     "in CUDA target architectures")
-    endif()
-  endif()
+  # The cutlass_scaled_mm kernels for Hopper (c3x, i.e. CUTLASS 3.x) require CUDA 12.0 or later
+  optional_cuda_sources(
+    NAME scaled_mm_c3x_sm90
+    MIN_VERSION 12.0
+    ARCHS "9.0a"
+    SRCS
+      "csrc/quantization/cutlass_w8a8/scaled_mm_c3x_sm90.cu"
+      "csrc/quantization/cutlass_w8a8/c3x/scaled_mm_sm90_fp8.cu"
+      "csrc/quantization/cutlass_w8a8/c3x/scaled_mm_sm90_int8.cu"
+      "csrc/quantization/cutlass_w8a8/c3x/scaled_mm_azp_sm90_int8.cu"
+      "csrc/quantization/cutlass_w8a8/c3x/scaled_mm_blockwise_sm90_fp8.cu"
+    FLAGS "-DENABLE_SCALED_MM_SM90=1"
+    APPEND_ARCHS SCALED_MM_3X_ARCHS
+    VERSION_MSG
+      "Not building scaled_mm_c3x_sm90: CUDA Compiler version is not >= 12.0.\n"
+      "Please upgrade to CUDA 12.0 or later to run FP8 quantized models on Hopper."
+  )
 
-  # The cutlass_scaled_mm kernels for Blackwell (c3x, i.e. CUTLASS 3.x) require
-  # CUDA 12.8 or later
-  cuda_archs_loose_intersection(SCALED_MM_ARCHS "10.0a;10.1a;12.0a" "${CUDA_ARCHS}")
-  if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER 12.8 AND SCALED_MM_ARCHS)
-    set(SRCS
+  # The cutlass_scaled_mm kernels for Blackwell (c3x, i.e. CUTLASS 3.x) require CUDA 12.8 or later
+  optional_cuda_sources(
+    NAME scaled_mm_c3x_sm100
+    MIN_VERSION 12.8
+    ARCHS "10.0a;10.1a;12.0a"
+    SRCS
       "csrc/quantization/cutlass_w8a8/scaled_mm_c3x_sm100.cu"
       "csrc/quantization/cutlass_w8a8/c3x/scaled_mm_sm100_fp8.cu"
       "csrc/quantization/cutlass_w8a8/c3x/scaled_mm_blockwise_sm100_fp8.cu"
-    )
-    set_gencode_flags_for_srcs(
-      SRCS "${SRCS}"
-      CUDA_ARCHS "${SCALED_MM_ARCHS}")
-    list(APPEND VLLM_EXT_SRC "${SRCS}")
-    list(APPEND VLLM_GPU_FLAGS "-DENABLE_SCALED_MM_SM100=1")
-    # Let scaled_mm_c2x know it doesn't need to build these arches
-    list(APPEND SCALED_MM_3X_ARCHS "${SCALED_MM_ARCHS}")
-    message(STATUS "Building scaled_mm_c3x_sm100 for archs: ${SCALED_MM_ARCHS}")
-  else()
-    if (NOT ${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER 12.8 AND SCALED_MM_ARCHS)
-      message(STATUS "Not building scaled_mm_c3x_sm100 as CUDA Compiler version is "
-                     "not >= 12.8, we recommend upgrading to CUDA 12.8 or "
-                     "later if you intend on running FP8 quantized models on "
-                     "Blackwell.")
-    else()
-      message(STATUS "Not building scaled_mm_c3x_100 as no compatible archs found "
-                     "in CUDA target architectures")
-    endif()
-  endif()
+    FLAGS "-DENABLE_SCALED_MM_SM100=1"
+    APPEND_ARCHS SCALED_MM_3X_ARCHS
+    VERSION_MSG
+      "Not building scaled_mm_c3x_sm100: CUDA Compiler version is not >= 12.8.\n"
+      "Please upgrade to CUDA 12.8 or later to run FP8 quantized models on Blackwell."
+  )
 
-  #
   # For the cutlass_scaled_mm kernels we want to build the c2x (CUTLASS 2.x)
   # kernels for the remaining archs that are not already built for 3x.
   # (Build 8.9 for FP8)
@@ -457,184 +378,87 @@ if(VLLM_GPU_LANG STREQUAL "CUDA")
     "7.5;8.0;8.9+PTX" "${CUDA_ARCHS}")
   # subtract out the archs that are already built for 3x
   list(REMOVE_ITEM SCALED_MM_2X_ARCHS ${SCALED_MM_3X_ARCHS})
-  if (SCALED_MM_2X_ARCHS)
-    set(SRCS "csrc/quantization/cutlass_w8a8/scaled_mm_c2x.cu")
-    set_gencode_flags_for_srcs(
-      SRCS "${SRCS}"
-      CUDA_ARCHS "${SCALED_MM_2X_ARCHS}")
-    list(APPEND VLLM_EXT_SRC "${SRCS}")
-    list(APPEND VLLM_GPU_FLAGS "-DENABLE_SCALED_MM_C2X=1")
-    message(STATUS "Building scaled_mm_c2x for archs: ${SCALED_MM_2X_ARCHS}")
-  else()
-    if (SCALED_MM_3X_ARCHS)
-      message(STATUS "Not building scaled_mm_c2x as all archs are already built"
-                     " for and covered by scaled_mm_c3x")
-    else()
-      message(STATUS "Not building scaled_mm_c2x as no compatible archs found "
-                    "in CUDA target architectures")
-    endif()
-  endif()
+  optional_cuda_sources(
+    NAME scaled_mm_c2x
+    ARCHS "${SCALED_MM_2X_ARCHS}"
+    SRCS "csrc/quantization/cutlass_w8a8/scaled_mm_c2x.cu"
+    FLAGS "-DENABLE_SCALED_MM_C2X=1"
+    NO_ARCH_MSG "Not building scaled_mm_c2x as no compatible archs found in CUDA target architectures, "
+                "or is already covered by scaled_mm_c3x."
+  )
 
   #
   # 2:4 Sparse Kernels
-
-  # The 2:4 sparse kernels cutlass_scaled_sparse_mm and cutlass_compressor
-  # require CUDA 12.2 or later (and only work on Hopper).
-  cuda_archs_loose_intersection(SCALED_MM_ARCHS "9.0a;" "${CUDA_ARCHS}")
-  if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER 12.2 AND SCALED_MM_ARCHS)
-    set(SRCS "csrc/sparse/cutlass/sparse_scaled_mm_c3x.cu")
-    set_gencode_flags_for_srcs(
-      SRCS "${SRCS}"
-      CUDA_ARCHS "${SCALED_MM_ARCHS}")
-    list(APPEND VLLM_EXT_SRC "${SRCS}")
-    list(APPEND VLLM_GPU_FLAGS "-DENABLE_SPARSE_SCALED_MM_C3X=1")
-    message(STATUS "Building sparse_scaled_mm_c3x for archs: ${SCALED_MM_ARCHS}")
-  else()
-    if (NOT ${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER 12.2 AND SCALED_MM_ARCHS)
-      message(STATUS "Not building sparse_scaled_mm_c3x kernels as CUDA Compiler version is "
-                     "not >= 12.2, we recommend upgrading to CUDA 12.2 or later "
-                     "if you intend on running FP8 sparse quantized models on Hopper.")
-    else()
-      message(STATUS "Not building sparse_scaled_mm_c3x as no compatible archs found "
-                     "in CUDA target architectures")
-    endif()
-  endif()
+  optional_cuda_sources(
+    NAME sparse_scaled_mm_c3x
+    MIN_VERSION 12.2
+    ARCHS "9.0a;"
+    SRCS "csrc/sparse/cutlass/sparse_scaled_mm_c3x.cu"
+    FLAGS "-DENABLE_SPARSE_SCALED_MM_C3X=1"
+    VERSION_MSG
+      "Not building sparse_scaled_mm_c3x: CUDA Compiler version is not >= 12.2.\n"
+      "Please upgrade to CUDA 12.2 or later to run FP8 sparse quantized models on Hopper."
+  )
 
   # FP4 Archs and flags
-  cuda_archs_loose_intersection(FP4_ARCHS "10.0a" "${CUDA_ARCHS}")
-  if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER 12.8 AND FP4_ARCHS)
-    set(SRCS
+  optional_cuda_sources(
+    NAME NVFP4
+    MIN_VERSION 12.8
+    ARCHS "10.0a"
+    SRCS
       "csrc/quantization/fp4/nvfp4_quant_kernels.cu"
       "csrc/quantization/fp4/nvfp4_experts_quant.cu"
       "csrc/quantization/fp4/nvfp4_scaled_mm_kernels.cu"
-      "csrc/quantization/fp4/nvfp4_blockwise_moe_kernel.cu")
-    set_gencode_flags_for_srcs(
-      SRCS "${SRCS}"
-      CUDA_ARCHS "${FP4_ARCHS}")
-    list(APPEND VLLM_EXT_SRC "${SRCS}")
-    list(APPEND VLLM_GPU_FLAGS "-DENABLE_NVFP4=1")
-    message(STATUS "Building NVFP4 for archs: ${FP4_ARCHS}")
-  else()
-    message(STATUS "Not building NVFP4 as no compatible archs were found.")
-    # clear FP4_ARCHS
-    set(FP4_ARCHS)
-  endif()
+      "csrc/quantization/fp4/nvfp4_blockwise_moe_kernel.cu"
+    FLAGS "-DENABLE_NVFP4=1"
+    NO_ARCH_MSG "Not building NVFP4 as no compatible archs were found."
+  )
 
   # CUTLASS MLA Archs and flags
-  cuda_archs_loose_intersection(MLA_ARCHS "10.0a" "${CUDA_ARCHS}")
-  if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER 12.8 AND MLA_ARCHS)
-    set(SRCS
-      "csrc/attention/mla/cutlass_mla_kernels.cu")
-    set_gencode_flags_for_srcs(
-      SRCS "${SRCS}"
-      CUDA_ARCHS "${MLA_ARCHS}")
-    list(APPEND VLLM_EXT_SRC "${SRCS}")
-    list(APPEND VLLM_GPU_FLAGS "-DENABLE_CUTLASS_MLA=1")
-    # Add MLA-specific include directories only to MLA source files
-    set_source_files_properties(${SRCS}
-      PROPERTIES INCLUDE_DIRECTORIES "${CUTLASS_DIR}/examples/77_blackwell_fmha;${CUTLASS_DIR}/examples/common")
-    message(STATUS "Building CUTLASS MLA for archs: ${MLA_ARCHS}")
-  else()
-    message(STATUS "Not building CUTLASS MLA as no compatible archs were found.")
-    # clear MLA_ARCHS
-    set(MLA_ARCHS)
-  endif()
+  optional_cuda_sources(
+    NAME CUTLASS_MLA
+    MIN_VERSION 12.8
+    ARCHS "10.0a"
+    SRCS "csrc/attention/mla/cutlass_mla_kernels.cu"
+    FLAGS "-DENABLE_CUTLASS_MLA=1"
+    NO_ARCH_MSG "Not building CUTLASS MLA as no compatible archs were found."
+  )
+  # Add MLA-specific include directories only to MLA source files
+  set_source_files_properties(
+    "csrc/attention/mla/cutlass_mla_kernels.cu"
+    PROPERTIES INCLUDE_DIRECTORIES "${CUTLASS_DIR}/examples/77_blackwell_fmha;${CUTLASS_DIR}/examples/common"
+  )
 
   # CUTLASS MoE kernels
-
-  # The MoE kernel cutlass_moe_mm requires CUDA 12.3 or later (and only works
-  # on Hopper). get_cutlass_moe_mm_data should only be compiled if it's possible
-  # to compile MoE kernels that use its output.
-  cuda_archs_loose_intersection(SCALED_MM_ARCHS "9.0a;10.0a" "${CUDA_ARCHS}")
-  if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.3 AND SCALED_MM_ARCHS)
-    set(SRCS "csrc/quantization/cutlass_w8a8/moe/grouped_mm_c3x.cu"
-             "csrc/quantization/cutlass_w8a8/moe/moe_data.cu")
-    set_gencode_flags_for_srcs(
-      SRCS "${SRCS}"
-      CUDA_ARCHS "${SCALED_MM_ARCHS}")
-    list(APPEND VLLM_EXT_SRC "${SRCS}")
-    list(APPEND VLLM_GPU_FLAGS "-DENABLE_CUTLASS_MOE_SM90=1")
-    message(STATUS "Building grouped_mm_c3x for archs: ${SCALED_MM_ARCHS}")
-  else()
-    if (NOT ${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER_EQUAL 12.3 AND SCALED_MM_ARCHS)
-      message(STATUS "Not building grouped_mm_c3x kernels as CUDA Compiler version is "
-                     "not >= 12.3, we recommend upgrading to CUDA 12.3 or later "
-                     "if you intend on running FP8 quantized MoE models on Hopper.")
-    else()
-      message(STATUS "Not building grouped_mm_c3x as no compatible archs found "
-                     "in CUDA target architectures")
-    endif()
-  endif()
+  optional_cuda_sources(
+    NAME grouped_mm_c3x
+    MIN_VERSION 12.3
+    ARCHS "9.0a;10.0a"
+    SRCS
+      "csrc/quantization/cutlass_w8a8/moe/grouped_mm_c3x.cu"
+      "csrc/quantization/cutlass_w8a8/moe/moe_data.cu"
+    FLAGS "-DENABLE_CUTLASS_MOE_SM90=1"
+    VERSION_MSG "Not building grouped_mm_c3x kernels as CUDA Compiler version is not >= 12.3, we recommend upgrading to CUDA 12.3 or later if you intend on running FP8 quantized MoE models on Hopper."
+  )
 
   #
   # Machete kernels
 
-  # The machete kernels only work on hopper and require CUDA 12.0 or later.
-  # Only build Machete kernels if we are building for something compatible with sm90a
+  # Machete kernels: generate and build for supported architectures
   cuda_archs_loose_intersection(MACHETE_ARCHS "9.0a" "${CUDA_ARCHS}")
-  if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER 12.0 AND MACHETE_ARCHS)
-    #
-    # For the Machete kernels we automatically generate sources for various
-    # preselected input type pairs and schedules.
-    # Generate sources:
-    set(MACHETE_GEN_SCRIPT
-      ${CMAKE_CURRENT_SOURCE_DIR}/csrc/quantization/machete/generate.py)
-    file(MD5 ${MACHETE_GEN_SCRIPT} MACHETE_GEN_SCRIPT_HASH)
-
-    message(STATUS "Machete generation script hash: ${MACHETE_GEN_SCRIPT_HASH}")
-    message(STATUS "Last run machete generate script hash: $CACHE{MACHETE_GEN_SCRIPT_HASH}")
-
-    if (NOT DEFINED CACHE{MACHETE_GEN_SCRIPT_HASH}
-        OR NOT $CACHE{MACHETE_GEN_SCRIPT_HASH} STREQUAL ${MACHETE_GEN_SCRIPT_HASH})
-      execute_process(
-        COMMAND ${CMAKE_COMMAND} -E env
-        PYTHONPATH=${CMAKE_CURRENT_SOURCE_DIR}/csrc/cutlass_extensions/:${CUTLASS_DIR}/python/:${VLLM_PYTHON_PATH}:$PYTHONPATH
-          ${Python_EXECUTABLE} ${MACHETE_GEN_SCRIPT}
-        RESULT_VARIABLE machete_generation_result
-        OUTPUT_VARIABLE machete_generation_output
-        OUTPUT_FILE ${CMAKE_CURRENT_BINARY_DIR}/machete_generation.log
-        ERROR_FILE ${CMAKE_CURRENT_BINARY_DIR}/machete_generation.log
-      )
-
-      if (NOT machete_generation_result EQUAL 0)
-        message(FATAL_ERROR "Machete generation failed."
-                            " Result: \"${machete_generation_result}\""
-                            "\nCheck the log for details: "
-                            "${CMAKE_CURRENT_BINARY_DIR}/machete_generation.log")
-      else()
-        set(MACHETE_GEN_SCRIPT_HASH ${MACHETE_GEN_SCRIPT_HASH}
-            CACHE STRING "Last run machete generate script hash" FORCE)
-        message(STATUS "Machete generation completed successfully.")
-      endif()
-    else()
-      message(STATUS "Machete generation script has not changed, skipping generation.")
-    endif()
-
-    # Add machete generated sources
-    file(GLOB MACHETE_GEN_SOURCES "csrc/quantization/machete/generated/*.cu")
-    list(APPEND VLLM_EXT_SRC ${MACHETE_GEN_SOURCES})
-
-    # forward compatible
-    set_gencode_flags_for_srcs(
-      SRCS "${MACHETE_GEN_SOURCES}"
-      CUDA_ARCHS "${MACHETE_ARCHS}")
-
-    list(APPEND VLLM_EXT_SRC
-      csrc/quantization/machete/machete_pytorch.cu)
-
-    message(STATUS "Building Machete kernels for archs: ${MACHETE_ARCHS}")
-  else()
-    if (NOT ${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER 12.0
-        AND MACHETE_ARCHS)
-      message(STATUS "Not building Machete kernels as CUDA Compiler version is "
-                     "not >= 12.0, we recommend upgrading to CUDA 12.0 or "
-                     "later if you intend on running w4a16 quantized models on "
-                     "Hopper.")
-    else()
-      message(STATUS "Not building Machete kernels as no compatible archs "
-                     "found in CUDA target architectures")
-    endif()
-  endif()
+  optional_cuda_sources(
+    NAME Machete
+    MIN_VERSION 12.0
+    ARCHS "${MACHETE_ARCHS}"
+    GEN_SCRIPT "${CMAKE_CURRENT_SOURCE_DIR}/csrc/quantization/machete/generate.py"
+    GEN_GLOB "csrc/quantization/machete/generated/*.cu"
+    SRCS "csrc/quantization/machete/machete_pytorch.cu"
+    VERSION_MSG
+      "Not building Machete kernels as CUDA Compiler version is less than 12.0."
+      "We recommend upgrading to CUDA 12.0 or later to run w4a16 quantized models on Hopper."
+    NO_ARCH_MSG
+      "Not building Machete kernels as no compatible archs found in CUDA target architectures"
+  )
 # if CUDA endif
 endif()
 
@@ -666,76 +490,22 @@ set(VLLM_MOE_EXT_SRC
   "csrc/moe/moe_align_sum_kernels.cu"
   "csrc/moe/topk_softmax_kernels.cu")
 
-if(VLLM_GPU_LANG STREQUAL "CUDA")
-  list(APPEND VLLM_MOE_EXT_SRC "csrc/moe/moe_wna16.cu")
-endif()
-
+# Apply gencode flags to base MOE extension sources
 set_gencode_flags_for_srcs(
   SRCS "${VLLM_MOE_EXT_SRC}"
   CUDA_ARCHS "${CUDA_ARCHS}")
 
+## Marlin MOE kernels: generate and include for supported architectures
 if(VLLM_GPU_LANG STREQUAL "CUDA")
-  set(VLLM_MOE_WNA16_SRC
-    "csrc/moe/moe_wna16.cu")
-
-  set_gencode_flags_for_srcs(
-    SRCS "${VLLM_MOE_WNA16_SRC}"
-    CUDA_ARCHS "${CUDA_ARCHS}")
-
-  list(APPEND VLLM_MOE_EXT_SRC "${VLLM_MOE_WNA16_SRC}")
-  # 9.0 for latest bf16 atomicAdd PTX
-  cuda_archs_loose_intersection(MARLIN_MOE_ARCHS "8.0;9.0+PTX" "${CUDA_ARCHS}")
-  if (MARLIN_MOE_ARCHS)
-
-    #
-    # For the Marlin MOE kernels we automatically generate sources for various
-    # preselected input type pairs and schedules.
-    # Generate sources:
-    set(MOE_MARLIN_GEN_SCRIPT
-      ${CMAKE_CURRENT_SOURCE_DIR}/csrc/moe/marlin_moe_wna16/generate_kernels.py)
-    file(MD5 ${MOE_MARLIN_GEN_SCRIPT} MOE_MARLIN_GEN_SCRIPT_HASH)
-
-    message(STATUS "Marlin MOE generation script hash: ${MOE_MARLIN_GEN_SCRIPT_HASH}")
-    message(STATUS "Last run Marlin MOE generate script hash: $CACHE{MOE_MARLIN_GEN_SCRIPT_HASH}")
-
-    if (NOT DEFINED CACHE{MOE_MARLIN_GEN_SCRIPT_HASH}
-        OR NOT $CACHE{MOE_MARLIN_GEN_SCRIPT_HASH} STREQUAL ${MOE_MARLIN_GEN_SCRIPT_HASH})
-      execute_process(
-        COMMAND ${CMAKE_COMMAND} -E env
-        PYTHONPATH=$PYTHONPATH
-          ${Python_EXECUTABLE} ${MOE_MARLIN_GEN_SCRIPT}
-        RESULT_VARIABLE moe_marlin_generation_result
-        OUTPUT_VARIABLE moe_marlin_generation_output
-        OUTPUT_FILE ${CMAKE_CURRENT_BINARY_DIR}/moe_marlin_generation.log
-        ERROR_FILE ${CMAKE_CURRENT_BINARY_DIR}/moe_marlin_generation.log
-      )
-
-      if (NOT moe_marlin_generation_result EQUAL 0)
-        message(FATAL_ERROR "Marlin MOE generation failed."
-                            " Result: \"${moe_marlin_generation_result}\""
-                            "\nCheck the log for details: "
-                            "${CMAKE_CURRENT_BINARY_DIR}/moe_marlin_generation.log")
-      else()
-        set(MOE_MARLIN_GEN_SCRIPT_HASH ${MOE_MARLIN_GEN_SCRIPT_HASH}
-            CACHE STRING "Last run Marlin MOE generate script hash" FORCE)
-        message(STATUS "Marlin MOE generation completed successfully.")
-      endif()
-    else()
-      message(STATUS "Marlin MOE generation script has not changed, skipping generation.")
-    endif()
-
-    file(GLOB MOE_WNAA16_MARLIN_SRC "csrc/moe/marlin_moe_wna16/*.cu")
-    set_gencode_flags_for_srcs(
-      SRCS "${MOE_WNAA16_MARLIN_SRC}"
-      CUDA_ARCHS "${MARLIN_MOE_ARCHS}")
-
-    list(APPEND VLLM_MOE_EXT_SRC ${MOE_WNAA16_MARLIN_SRC})
-
-    message(STATUS "Building Marlin MOE kernels for archs: ${MARLIN_MOE_ARCHS}")
-  else()
-    message(STATUS "Not building Marlin MOE kernels as no compatible archs found"
-                   " in CUDA target architectures")
-  endif()
+  optional_cuda_sources(
+    NAME "Marlin MOE"
+    ARCHS "8.0;9.0+PTX"
+    GEN_SCRIPT "${CMAKE_CURRENT_SOURCE_DIR}/csrc/moe/marlin_moe_wna16/generate_kernels.py"
+    GEN_GLOB "csrc/moe/marlin_moe_wna16/*.cu"
+    SRCS "csrc/moe/moe_wna16.cu"
+    NO_ARCH_MSG "Not building Marlin MOE kernels as no compatible archs found in CUDA target architectures"
+    OUT_SRCS_VAR VLLM_MOE_EXT_SRC
+  )
 endif()
 
 if(VLLM_GPU_LANG STREQUAL "CUDA")
diff --git a/cmake/utils.cmake b/cmake/utils.cmake
index 12e4e39024f5d..2ef0eb43897da 100644
--- a/cmake/utils.cmake
+++ b/cmake/utils.cmake
@@ -39,6 +39,34 @@ function (run_python OUT EXPR ERR_MSG)
   set(${OUT} ${PYTHON_OUT} PARENT_SCOPE)
 endfunction()
 
+# Generate CUDA sources via a Python script with caching based on script hash
+function(generate_cuda_sources NAME SCRIPT GLOB OUT_SRCS)
+  string(TOUPPER "${NAME}" _UPPER_NAME)
+  set(_CACHE_VAR "${_UPPER_NAME}_GEN_SCRIPT_HASH")
+  file(MD5 "${SCRIPT}" _GEN_HASH)
+  message(STATUS "${NAME} generation script hash: ${_GEN_HASH}")
+  message(STATUS "Last run ${NAME} generation script hash: $CACHE{${_CACHE_VAR}}")
+  if(NOT DEFINED CACHE{${_CACHE_VAR}} OR NOT $CACHE{${_CACHE_VAR}} STREQUAL "${_GEN_HASH}")
+    execute_process(
+      COMMAND ${CMAKE_COMMAND} -E env
+        PYTHONPATH=$ENV{PYTHONPATH}
+        ${Python_EXECUTABLE} "${SCRIPT}"
+      RESULT_VARIABLE _GEN_RESULT
+      OUTPUT_FILE "${CMAKE_CURRENT_BINARY_DIR}/${NAME}_generation.log"
+      ERROR_FILE "${CMAKE_CURRENT_BINARY_DIR}/${NAME}_generation.log"
+    )
+    if(NOT _GEN_RESULT EQUAL 0)
+      message(FATAL_ERROR "${NAME} generation failed. Result: \"${_GEN_RESULT}\"\nCheck the log for details: ${CMAKE_CURRENT_BINARY_DIR}/${NAME}_generation.log")
+    else()
+      set(${_CACHE_VAR} "${_GEN_HASH}" CACHE STRING "Last run ${NAME} generation script hash" FORCE)
+      message(STATUS "${NAME} generation completed successfully.")
+    endif()
+  else()
+    message(STATUS "${NAME} generation script has not changed, skipping generation.")
+  endif()
+  file(GLOB ${OUT_SRCS} "${GLOB}")
+endfunction()
+
 # Run `EXPR` in python after importing `PKG`. Use the result of this to extend
 # `CMAKE_PREFIX_PATH` so the torch cmake configuration can be imported.
 macro (append_cmake_prefix_path PKG EXPR)
@@ -86,6 +114,78 @@ function (hipify_sources_target OUT_SRCS NAME ORIG_SRCS)
   set(${OUT_SRCS} ${HIP_SRCS} PARENT_SCOPE)
 endfunction()
 
+## Macro to conditionally include CUDA sources based on architecture and CUDA
+## compiler version, optionally generating sources via a Python script.
+## Usage:
+##   optional_cuda_sources(
+##     NAME <name>
+##     [MIN_VERSION <version>]
+##     ARCHS <arch1;arch2;...>
+##     SRCS <static_source1> [<static_source2> ...]
+##     [FLAGS <flag1> ...]
+##     [VERSION_MSG <line1> [<line2> ...]]
+##     [NO_ARCH_MSG <line1> [<line2> ...]]
+##     [GEN_SCRIPT <path/to/generate_script.py>]
+##     [GEN_GLOB <glob_pattern_for_generated_sources>]
+##     [APPEND_ARCHS <var_to_append_archs>]
+## This will run GEN_SCRIPT once when version and arch checks pass, globbing
+## sources matching GEN_GLOB and appending them alongside SRCS.
+macro(optional_cuda_sources)
+  set(oneValueArgs NAME MIN_VERSION APPEND_ARCHS GEN_SCRIPT GEN_GLOB OUT_SRCS_VAR)
+  set(multiValueArgs ARCHS SRCS FLAGS VERSION_MSG NO_ARCH_MSG)
+  cmake_parse_arguments(OCS "" "${oneValueArgs}" "${multiValueArgs}" ${ARGN})
+  if(NOT OCS_NAME)
+    message(FATAL_ERROR "optional_cuda_sources: NAME is required")
+  endif()
+  if(NOT OCS_ARCHS)
+    message(FATAL_ERROR "optional_cuda_sources ${OCS_NAME}: ARCHS is required")
+  endif()
+  if(NOT OCS_SRCS)
+    message(FATAL_ERROR "optional_cuda_sources ${OCS_NAME}: SRCS is required")
+  endif()
+  if(NOT OCS_MIN_VERSION)
+    set(OCS_MIN_VERSION "0.0")
+  endif()
+  cuda_archs_loose_intersection(_OCS_ARCHS "${OCS_ARCHS}" "${CUDA_ARCHS}")
+  if(CMAKE_CUDA_COMPILER_VERSION VERSION_GREATER_EQUAL ${OCS_MIN_VERSION})
+    if(_OCS_ARCHS)
+      set(_OCS_SRCS ${OCS_SRCS})
+      # Generate sources if a script is provided
+      if(OCS_GEN_SCRIPT AND OCS_GEN_GLOB)
+        generate_cuda_sources(${OCS_NAME} "${OCS_GEN_SCRIPT}" "${OCS_GEN_GLOB}" _OCS_GEN_SRCS)
+        list(APPEND _OCS_SRCS ${_OCS_GEN_SRCS})
+      endif()
+      set_gencode_flags_for_srcs(SRCS "${_OCS_SRCS}" CUDA_ARCHS "${_OCS_ARCHS}")
+      if(OCS_OUT_SRCS_VAR)
+        list(APPEND ${OCS_OUT_SRCS_VAR} ${_OCS_SRCS})
+      else()
+        list(APPEND VLLM_EXT_SRC ${_OCS_SRCS})
+      endif()
+      if(OCS_FLAGS)
+        list(APPEND VLLM_GPU_FLAGS ${OCS_FLAGS})
+      endif()
+      if(OCS_APPEND_ARCHS)
+        list(APPEND ${OCS_APPEND_ARCHS} ${_OCS_ARCHS})
+      endif()
+      message(STATUS "Building ${OCS_NAME} for archs: ${_OCS_ARCHS}")
+    else()
+      if(OCS_NO_ARCH_MSG)
+        list(JOIN OCS_NO_ARCH_MSG "\n" _OCS_NO_ARCH_JOINED)
+        message(STATUS "${_OCS_NO_ARCH_JOINED}")
+      else()
+        message(STATUS "Not building ${OCS_NAME}: no compatible architectures found in CUDA target architectures")
+      endif()
+    endif()
+  else()
+    if(OCS_VERSION_MSG)
+      list(JOIN OCS_VERSION_MSG "\n" _OCS_VERSION_JOINED)
+      message(STATUS "${_OCS_VERSION_JOINED}")
+    else()
+      message(STATUS "Not building ${OCS_NAME}: CUDA Compiler version is less than ${OCS_MIN_VERSION}")
+    endif()
+  endif()
+endmacro()
+
 #
 # Get additional GPU compiler flags from torch.
 #