using CMake to turn CUDA compilation on or off for only a sub directory

So, in the project I’m currently involved with, it looks like this:

project
    |--CMakeLists.txt 
    |--app
    |   |--CMakeLists.txt
    |   |--app_to_cuda
    |       |--CMakeLists.txt
    |       |--app_to_cuda.cc
    |       |--app_to_cuda.h
    |--src
       |--CMakeLists.txt
       |--<source_files>

Inside project/app/CMakeLists.txt, I have done the following:

add_subdirectory(app_to_cuda)

And then inside project/app/app_to_cuda/CMakeLists.txt:

add_executable(app_to_cuda app_to_cuda.cc)
    target_link_libraries(app_to_cuda ${SRC_LIBS} ${OTHER_LIBS})

Now, I want to introduce a new file in project/app/app_to_cuda to allow CUDA porting, if possible, only for app_to_cuda. To be specific, I’m changing the source tree into this:

project
    |--CMakeLists.txt 
    |--app
    |   |--CMakeLists.txt
    |   |--app_to_cuda
    |       |--CMakeLists.txt
    |       |--app_to_cuda.cc
    |       |--app_to_cuda.h
    |       |--app_to_cuda.cu <--- Written with CUDA Kernel and stuff
    |--src
       |--CMakeLists.txt
       |--<source_files>

So in order to do this, I have edited project/app/app_to_cuda/CMakeLists.txt into:

if (ENABLE_CUDA_APP)
  find_package(CUDA REQUIRED)
  include(FindCUDA)

  list(APPEND CUDA_NVCC_FLAGS -gencode arch=compute_70,code=sm_70)

  cuda_add_executable(app_to_cuda app_to_cuda.cu)
  target_link_libraries(app_to_cuda ${OTHER_LIBS})
else (ENABLE_CUDA_APP)
  add_executable(app_to_cuda app_to_cuda.cc)
  target_link_libraries(app_to_cuda ${OTHER_LIBS})
endif (ENABLE_CUDA_APP)

But it doesn’t seem to be working. First, it complains about the warning:

[...]
warning: "THRUST_CUB_NS_POSTFIX" redefined
 #define THRUST_CUB_NS_POSTFIX }  }
[...]

And then the error:

[...]
error: statement may not appear in a constexpr function
[...]
error: a constexpr function must contain exactly one return statement
[...]

I’m kinda lost at this point. Any suggestions on how I could resolve this?