Accurate depth estimation is crucial for safe perception of autonomous / (highly) automated vehicles. Monocular Depth Estimation (MDE) techniques are gaining traction in the automotive sector due to their cost-effectiveness and ease of integration, making them promising for several downstream tasks in the automated driving domain. However, most current state-of-the-art (SOTA) MDE models rely on computationally heavy architectures and are evaluated mainly on benchmark datasets, leaving their generalisation ability and suitability for resource-constrained edge hardware uncertain. This thesis evaluates the feasibility of deploying MDE models in automated vehicles through two objectives. First, it examines the generalisation ability and resource demands of current SOTA MDE models by evaluating them on previously unseen datasets, analysing the trade-off between accuracy and efficiency. Second, it investigates whether responsebased knowledge distillation can produce a student model that approaches the accuracy of a high-performing teacher while remaining lightweight enough for deployment on edge hardware and meeting the operational requirements of automated vehicles.