In the spirit of this library, we want more code is as reasonable to exist on the Rust side versus the C++ side. To that end, it would greatly enhance flexibility if we could pass device pointers to libinfer in order to execute operations. This will allow chained CUDA operations to be feasible without additional host to device memcpys.