Small Renal Masses (SRM), (renal masses ≤ 4.0 cm in diameter), present significant diagnostic challenges when using radiological images. Current contrast enhancement-based methods need further advancements in fine subclassification of SRMs to identify the need to rule out borderline malignancy, and the need for regular monitoring. In this study, we propose an enhanced deep learning model that integrates convolutional layers, channel expansion and squeeze to extract important features, cross-channel attention, selective channel retention and residual links for an automated subclassification of T1a SRMs to improve diagnostics and medical treatment by ruling out possible future malignancy. Convolutional layers extract local features. Channel expansion and squeeze, augmented with cross-channel attention and selective channel retention, enhance the selection of the most informative features. Residual links help mitigate information-loss during convolutional transformation. We present an algorithm and evaluate our model on the KiTS19 and KiTS21 challenge datasets, comprising 8,262 CT slices. The model achieved a validation accuracy of 98.2% and a test accuracy of 98.5% in identifying healthy kidneys and subclassifying T1a tumors into three sub-categories: 1.2–2.0 cm (T1a1), 2.0–3.0 cm (T1a2), and 3.0–4.0 cm (T1a3). Our Model outperforms other related models.