Hey @Rahul_Reddy,
Welcome to the community. Before I write anything, I would like to highlight that I am not an AI professional, but as an AI learner, let me give you my 2 cents.
Hardly any one will ask you to write a 5-layer or 15-layer NN from scratch, as we have amazing frameworks for that in place as @alvaroramajo pointed it out very well. But most certainly you can be asked to develop a new CNN model, for instance in one of your research projects. Even in that case, you can use these frameworks to write your CNN very efficiently. For instance, check this repo out.
In this case, I had to write my custom loss function and my custom metric, but tensorflow makes it extremely easy, especially with the auto-grad feature, and so does PyTorch. You can even write your custom layers with just a few lines of code, and voila, you are done.
Now, as to feeling completely overwhelmed by the code, it’s perfectly natural for a beginner to feel that way. I myself felt that way when I was starting out with AI, but now, I love to go through the source code of Tensorflow (PyTorch a bit less 
), and sometimes, I end up spending hours going through the source code, just to figure out something, so, just keep at it, and perhaps, you could come up with a framework of your own someday, that all of us will be learning in the DLS.
Also, there are some tips from Prof Andrew about launching a career in AI, which you can find here. This is a bit unrelated to your topic, but I thought this could help you.
Cheers,
Elemento