Neural networks are now a key element of many complex software systems, typically known as AI-enhanced systems or smart systems. Model-driven engineering (MDE) approaches are often used to model and (semi)automatically generate such complex systems. Nevertheless, for MDE to be used in the creation of smart systems, it is essential to be able to model the implementation of neural networks (NNs) as part of the rest of the system. Unfortunately, modelling support for NNs is rather limited. In particular, there is a lack of complete and expressive neural network metamodels that could facilitate NN development. This poses significant challenges as it creates a disconnect between the development of neural networks and the other components of the system. Overall, this paper introduces a metamodel and its concrete syntax to support neural network design, providing a foundation for future research in smart system development.