This article introduces an innovative solution to critical challenges in the automated detection of Diabetic Retinopathy using fundus images. To combat data scarcity, we investigate the process of modeling fundus images by proposing the 'Retina-Based Affine Mapping' mechanism. This facilitated the generation of representative augmentations to model occurrences influenced by various internal and external factors during fundus image acquisition, diverging away from the concept of previous works focusing on generic augmentations focused primarily on data scaling rather than increased representation. Additionally, we propose a 'Design Flow Mechanism' to streamline custom Convolutional Neural Network architecture development, via an internal parameter comparison table, resulting in a highly efficient model with 99.51% validation accuracy using only 1.40 million parameters, outperforming state-of-the-art alternatives such as ResNet-18, consisting of 11.69 million learnable parameters. These contributions enhance the field of automated DR detection, promising significant advancements in medical image analysis and early disease diagnosis.