The rapid pace of urbanization and the growing number of vehicles has driven the demand for advanced parking management systems. However, a significant challenge remains in maximizing the efficiency of parking space utilization. While deep learning algorithms have proven effective in detecting available parking spots, their accuracy can be compromised by obstacles such as partial obstructions and varying lighting conditions. These issues are particularly pronounced under foggy or hazy weather, leading to a noticeable drop in performance. This research article presents an innovative approach that combines the strengths of Convolutional Neural Networks (CNN) with Kernel Extreme Learning Machine (KELM) to improve accuracy and reduce training time in parking space classification. The core advancement lies in substituting the fully connected layer of the CNN with KELM, which eliminates the need for backpropagation, thereby cutting down the training duration. The proposed approach is validated on the PKLot dataset, and a custom-built dataset designed to classify parking spaces under hazy conditions is used. The images from this hazy dataset are first preprocessed with Light-DehazeNet (LD-Net) to mitigate the impact of haze. Then, the CNN is trained on the dehazed dataset to extract features and produce a feature vector. This vector is then fed into the KELM to classify the parking spaces as occupied or vacant. Compared to standalone CNN, KELM, other machine learning classifiers, and similar models, the proposed approach demonstrates superior performance and enhanced computational efficiency. The study illustrates that the combination of CNN and KELM offers a promising solution for accurately classifying parking spaces, even in challenging weather conditions.