Computer-aided diagnosis of diseases has proven to be a major field that has benefited from the rapid advancements in artificial intelligence and its branches of machine learning and deep learning. Machine and deep learning-based models have found vast applications in solving biomedical classification problems, with Ocular Diseases proving no different. An early diagnosis of eye diseases can prevent patients from permanently losing their vision. Still, human diagnosis of the same has proven to be too complex a task to reach the masses. Therefore, recent work has been guided to deploy several Convolutional Neural Networks (CNNs) based ensemble DL models to assist clinicians in assisting in Ophthalmic Disease (OD) detection. This work takes a visual dataset consisting of fundus images and proposes a self-attention integrated EfficientNetV2B3 model that reads the required characteristics from them for correctly identifying the Ocular disease. The work also presents a detailed analysis of the optimal placement of the required number of Self-attention layers in the architecture without significantly increasing the model complexity. For comprehensive analysis, the work also implements and analyzes five recent state-of-art models namely, EfficientNet V2B3, DenseNet201, InceptionV3, Resnet50, and Xception, and studies their efficiency in correctly dealing with the multi-class disease detection at hand. The proposed model’s performance has been found to be the best and surpasses the other state-of-art models by 4% in terms of F-score, 5% in terms of Recall and 8% in terms of accuracy with its F-score currently at 64%. It is considered to be a good value as the dataset contains images having eight classes with slight variations in inter-class images.