Recently, surging urbanization and the subsequent increase in vehicular movement have amplified the popularity of parking management systems. A crucial hurdle in effective parking management is the optimal utilization of available parking spaces. Although deep learning techniques have showcased their prowess in accurately identifying parking spaces, they sometimes falter when partial obstructions and diverse lighting conditions are present. This challenge is further exacerbated during foggy or hazy weather, resulting in a decline in their performance. This research article introduces a novel approach that leverages the benefits of CNN and Kernel Extreme Learning Machine (KELM). The approach not only enhances accuracy but also significantly reduces training time. The key innovation lies in replacing the fully connected layer of CNN with KELM, which circumvents the backpropagation process and reduces the training time. A purpose-built data set tailored to classify parking spaces in hazy conditions and PKLot is employed for the validation. Within the proposed approach, images from the hazy dataset undergo preprocessing using Light-DehazeNet (LD-Net) to counteract the effects of haze. Subsequently, CNN is trained on this preprocessed dataset to extract features and generate a feature vector. This feature vector is input into the KELM to classify parking spaces as occupied or vacant. Compared with standalone CNN, KELM, machine learning classifiers, and similar existing models, the proposed model showcases superior performance and better computational efficiency. This study demonstrates that the integration of CNN with KELM offers a promising solution for parking space classification in challenging weather conditions.