Despite the modern advances in the available computational resources, the length and time scales of the physical systems that can be studied in full atomic detail, via molecular simulations, are still limited. To overcome such limitations, methods based on hierarchical multi-scale modelling have been developed. Hierarchical multi-scale modelling involves multiple scales of description and offers a powerful concept for tackling large systems and protracted equilibration times. In recent years, deep learning has emerged as a powerful tool for addressing the complexities of representing such high-dimensional functions. This breakthrough offers an unprecedented opportunity to revisit and enhance the theoretical foundations of various scientific fields, develop innovative methodologies, and solve problems that were previously too complex for traditional approaches. This thesis explores several of these challenges within the domain of multi-scale modeling. We propose deep learning-based techniques that significantly enhance simulation accuracy and efficiency, surpassing the capabilities of conventional methods. We focus on molecular modeling, employing theoretical methods and computational techniques to simulate and study molecular behavior, from small chemical systems to large multi-component molecular systems and material assemblies. The primary objective of this dissertation is to systematically demonstrate these methods and highlight the significant applications enabled by these new tools, thereby showcasing the transformative potential of deep learning in advancing multi-scale modeling.