Peripheral neuropathies represent a significant medical challenge, severely impacting health and quality of life. This study leverages a Multiple Input Multiple Output (MIMO) model to classify electroneurographic (ENG) signals recorded via multi-contact cuff electrodes. By optimizing neural network architectures, including the Electroneurographic Network (ENGNet) and Inception Transformer (IT), the study demonstrates superior classification performance, with ENGNet achieving the highest accuracy and robustness. An extensive ablation study offers detailed insights into the roles of individual network components. These results pave the way for integrating these models into neuroprosthetic systems, potentially enabling real-time sensory and motor restoration with high precision and efficiency.