Deep learning models have witnessed immense empirical success over the last decade. However, in spite of their widespread adoption, a profound understanding of the generalization behaviour of these over-parameterized architectures is still missing. In this thesis, we provide one such way via a data-dependent characterizations of the generalization capability of deep neural networks based data representations. In particular, by building on the algorithmic robustness framework, we offer a generalisation error bound that encapsulates key ingredients associated with the learning problem such as the complexity of the data space, the cardinality of the training set, and the Lipschitz properties of a deep neural network. We then specialize our analysis to a specific class of model based regression problems, namely the inverse problems. These problems often come with well defined forward operators that map variables of interest to the observations. It is therefore natural to ask whether such knowledge of the forward operator can be exploited in deep learning approaches increasingly used to solve inverse problems. We offer a generalisation error bound that -- apart from the other factors -- depends on the Jacobian of the composition of the forward operator with the neural network. Motivated by our analysis, we then propose a `plug-and-play' regulariser that leverages the knowledge of the forward map to improve the generalization of the network. We likewise also provide a method allowing us to tightly upper bound the norms of the Jacobians of the relevant operators that is much more {computationally} efficient than existing ones. We demonstrate the efficacy of our model-aware regularised deep learning algorithms against other state-of-the-art approaches on inverse problems involving various sub-sampling operators such as those used in classical compressed sensing setup and inverse problems that are of interest in the biomedical imaging setup.