Diabetic retinopathy is extensively accepted as a important etiology of visual impairment on a global scale. These particular ailment impacts individuals afflicted with diabetes, particularly those who are advanced in age. Numerous hospitals worldwide endeavor to engage in research and implement preventive measures aimed at mitigating the incidence of blindness resulting from diabetic retinopathy. Cataract, Diabetic Retinopathy, and Glaucoma are prevalent eye illnesses that are frequently encountered on a global scale. The suggested model has undergone training using a substantial dataset of fundus images in order to accomplish several tasks concurrently. These tasks encompass diabetic retinopathy (DR) categorization, Gaussian blur, and grey scaling. Through the process of combining all tasks, the algorithm is capable of generating predictions that are more informed and enhancing its overall efficacy in contrast to conventional single-task models. One of the benefits of utilizing Convolutional Neural Networks (CNN) and Transfer Learning (TL) methodologies is their ability to enhance accuracy. A dataset of 4000 photos was compiled from publicly accessible sources, namely Ocular Recognition, HRF, DRIVE, and IDRiD. This dataset was then validated using a subset of 402 images, which were selected from a larger dataset comprising 1004 images for each of the classes. The validation process resulted in a precision of 90.89%.