Firstly, data preprocessing is a necesaary step to any machine learning work. The reason for this is that during the collection and sometimes transfer of data (to and from data storage location like a server or local disk),discrepancies are introduced in the data. Since machine learning requires that your data be uniform, it is imperative to conduct some form of data preprocessing to get them in the right format for your work. Let me also mention that there is no clear cut formula to it as the kind of preprocessing will depend on the task. Some tasks require heavy preprocessing while others will do just fine with the bare minimum of preprocessing.
Secondly, what you have listed are not models but individual layers that make up your model. Depending on your tasks, the layers will be stacked together in a particular order to achieve the performance you desire (although other things also come in play here). So just as mentioned in the first point, your choice of data preprocessing still boils down to your tasks.
The tutorial section of the tensorflow docs has great resource for practising and learning.You can give that a try first.
