Unless one performs an exhaustive search, it's difficult to answer your question.
However, in the widely used libraries, such as TensorFlow, PyTorch and sklearn, most abstractions (like neural networks and layers) are implemented as classes (see this, this and this examples, respectively), as the main programming language supported by these libraries is Python, which an object-oriented language (but note that Python also supports other programming paradigms, such as functional programming).
I don't know the statistics, but, from my experience, I would say that OOP (which tends to be intuitive to humans for obvious reasons) is the mostly widely used programming paradigm, as opposed to the (pure) functional paradigm.
However, in general, the programming paradigm used to implement a certain concept probably depends on the language that you want to use. For example, in Haskell, a purely functional programming language, you will probably implement a perceptron as a sequence of functions (see this example). Another example is NumPy, which, although the primary interface is written in Python, under the hood, is primarily implemented in C, a non-OOP language (see e.g. this example, where you see many functions, but no class).
This should also partially answer your other question. In some cases, you will implement a concept using the programming language and paradigm that improves the efficiency of your implementation (e.g. the case of NumPy).