Embedding pytorch for machine learning iusing DSP

Dear PyTorch development community,
We are developing an application of neural network to run on an embedded device with microprocessor + DSP. Our application will perform the learning process so we need to perform the forward and backward propagation on the embedded device. We are assessing the possibility of using Pytorch and would like to know from you if your experts have any suggestions regarding this task and/or if there is any kind of incompatibility that we can face due to the use o DSPs with the library.