CMake compute model

[Aleksei91]

Hello, I'm building YOLO for Windows 10 via CMake. I think there should be an option for choice multiple compute models (nvcc -gencode arch=...,code=... options), like when I'm building OpenCV there is CUDA_ARCH_BIN option with multiple versions. Is there any way to do it now? Is there any reason for single architecture in CUDA_COMPUTE_MODEL option?