<p>Efficient predictive models can contribute to derive improved energy management systems for electric direct-drive wheels. This paper presents Deep Gaussian Processes (DGPs) as the extension of traditional GPs based on additional layered architectures. A controlled lab experiment with a quarter-scale vehicular setup was used to acquire experimental data needed to develop the predictive model. The experimental setup simulates a commercial vehicle driving environment and measures force dynamics converted to energy consumed by the electric direct-drive wheels. In order to develop the predictive model, the higher-order interface systems were first introduced to consider the nonlinear interconnections between GP layers and improve uncertainty propagation along with complex data. A fundamental characteristic of this paper is the theoretical exploration of DGPs based on a methodical technique for adding layers within the DGP framework. This approach helps to manage complex energy consumption predictions in electric commercial vehicles. The inherent properties of DGPs, which emphasize their compositionality are examined. At the same time, the combination of multiple GPs that remain Gaussian are also assessed. Finally, the convergence patterns of the marginal likelihood in deep structures are demonstrated, which defines conditions for its stability and guarantees the model’s adaptability to various data complexities.</p>