Your need to work the representtion of each task. For example, Naive Bayes you need to understand the Graphical probabilistic graphical model behind. Then, deducing the loss function from the MAP inference problem is straightforward. A well established reference in this manner is the Bishop book "Pattern Recognitin and machine learning"

Here is the link:

http://users.isr.ist.utl.pt/~wurmd/Livros/school/Bishop%20-%20Pattern%20Recognition%20And%20Machine%20Learning%20-%20Springer%20%202006.pdf