Remove multiprosessing

We want the sytem to estimate faster. Removing the multiprocessing will make the API constantly occupy the GPU, but hopefully it will estimate faster because it doesn't have to implement tensorflow before every estimation. Hopefully there is no memory leak.