Due to noteworthy advancements in hardware computing capabilities, Convolutional Neural Networks (CNNs) in deep learning (DL) techniques have been showing remarkable performance in terms of fast analysis, precision, cost-effectiveness and user-friendliness in the field of medical science. Though, CNN architectures pose challenges when it comes to utilizing them with inadequate technological infrastructure in rural areas. These challenges inspired us to propose a Tiny InceptionResNetV2 Model for diabetic retinopathy (DR) diagnosis using retinal images. Initially, the visual quality of the retinal images is enhanced through numerous imaging techniques. Then the proposed model extracts the features from retinal images. Finally, the output features are applied as the input to a machine learning (ML) algorithm for classification purpose. The efficacy of the proposed approach is assessed through K-fold cross-validation technique, which also helps to prevent over-fitting in the system. The K-Nearest Neighbor (KNN) classifier has ensured promising outcomes with the highest accuracy of 98.09% for multi-class classification. In light of the experimental findings, the suggested Tiny InceptionResNetV2 Model holds potential for real-world use, making it possible to evaluate DR classification even in the most technologically under-developed areas.