Mixing implicit and explicit CUDA streams

Question

I have some CUDA 8.0 code that looks basically like this:

cudaMemcpy(devInputData, ..., cudaMemcpyHostToDevice);
kernelThings<<