You can't define a model "NPA" in which all numbers are nonpositive but then define multiplication such that it can yield positive numbers - that just wouldn't be a function.
It's pretty easy to prove that S(0)*S(0) = S(0) with the axioms of PA. That has to be true no matter what interpretation you give "S" in your model, so in your proposed model "NPA", in which the domain are the nonpositive integers and S is the predecessor function, of course it has to be true that the "product" of two -1 and -1 is -1. It just means that in such a model, multiplication isn't what you think it is.
One should really be careful not to mix up logical systems and their models. The way you're using notation, it's really easy to get confused.
It's pretty easy to prove that S(0)*S(0) = S(0) with the axioms of PA. That has to be true no matter what interpretation you give "S" in your model, so in your proposed model "NPA", in which the domain are the nonpositive integers and S is the predecessor function, of course it has to be true that the "product" of two -1 and -1 is -1. It just means that in such a model, multiplication isn't what you think it is.
One should really be careful not to mix up logical systems and their models. The way you're using notation, it's really easy to get confused.