Numerous eye disorders have demonstrated encouraging improvements with the development of image processing techniques and deep learning (DL) approaches. However, many of these studies target on a single disease. Consequently, it is effective to concentrate on multi-disease classification utilizing retinal fundus images. This study is to examine the role of image processing techniques in the classifications of retinal diseases using CNN models like VGG19, ResNet50 and SqueezNet. The performance indicators, including accuracy, F1 score, recall, and precision, are used to evaluate the model's performance with and without image processing techniques for retinal diseases classification. The results shows that the model’s efficiency is improved by employing appropriate image processing techniques before fed into the classifier. The outcome of this study is the development of a reliable and efficient diagnostic system for identifying and treating of several eye diseases using color retinal image analysis.