Server architecture


I have a question about the design of my infrastructure. I have a website which receives images and annotations from users. With these images I want to train a CNN. What I currently have in mind is:

The masterserver that receives the images will send them in batches of 100 to the CNN-server. The CNN-server has a python api framework installed to handle these requests. After the images have been transferred the CNN will start training with these images and delete them after training is complete.

Is this a proper way to set-up my infrastructre or are there better options?