Here comes an experimentalist point of view.These models are approximations of the real theoretical description of matter in aggregate,which at the moment is not a simple one.
To truly describe nuclear forces in the language of QCD , our current knowledge of the theory of strong interactions, is a formidable many body task.
In the calculational methods that exist now you would need thousands of Feynman diagrams calculated for each specific question. The models, fitted to real data, allow one to describe the behavior of nuclei and have some predictive power over it.
Qualitatively one can think of the drop model as an approximation to the continuity of force interchanges with gluon exchanges between the quarks of the nuclei within the nuclear bag. At high energies one talks of quark-gluon plasma. This is the liquid phase analogy.
The shell model works as a total approximation to a potential well, similar to the way the Bohr model worked for describing the atom. There is a collective potential well there, created by all the gluon exchanges, and it is approximated in this simplified form using data to be able to predict further data. It works.