The diagnosis of skin disease is a critical area in healthcare, requiring high accuracy and reliability to ensure effective treatment. Traditional diagnostic methods often struggle with the variability in the appearance of skin lesions and noise in medical images, leading to misdiagnosis. This study addresses these challenges by proposing a robust deep learning-based framework for accurate skin disease classification. Motivated by the need for precise and automated diagnostic tools, we focus on enhancing image quality and leveraging advanced convolutional neural networks (CNN) to improve classification performance. The methodology involves comprehensive data preprocessing, including image resizing, morphological black hat transformation, median filtering, and contrast adjustment to highlight fine details and reduce noise. Data augmentation techniques such as flipping, rotating, scaling, and shifting are employed to increase dataset diversity. A systematic evaluation of state-of-the-art and custom CNN architectures is conducted, with rigorous hyperparameter tuning to optimize performance. Our results demonstrate exceptional performance, achieving 98.57% accuracy, high specificity, precision, recall, and F1 score, supported by AUC-ROC analysis and 5-fold cross-validation. The proposed model outperforms existing architectures, showcasing its potential for precise diagnosis of skin disease. This study highlights the effectiveness of combining advanced preprocessing techniques with deep learning models to address the complexities of skin disease classification, paving the way for reliable automated diagnostic systems.