Diabetic Retinopathy (DR) is a commonly occurring disease among diabetic patients that affects retina lesions and vision. Since DR is irreversible, an earlier diagnosis of DR can considerably decrease the risk of vision loss. Manual detection and classification of DR from retinal fundus images is time-consuming, expensive, and prone to errors, contrasting to CAD models. In recent times, DL models have become a familiar topic in several applications, particularly medical image classification. With this motivation, this paper presents new deep learning-empowered diabetic retinopathy detection and classification (DL-DRDC) model. The DL-DRDC technique aims to recognize and categorize different grades of DR using retinal fundus images. The proposed model involves the Contrast Limited Adaptive Histogram Equalization (CLAHE) technique as a pre-processing stage, which is used to enhance the contrast of the fundus images and improve the low contrast of medical images. Besides, the CLAHE is applied to the L channel of the retina images that have higher contrast. In addition, a deep learning-based Efficient Net-based feature extractor is used to generate feature vectors from pre-processed images. Moreover, a deep neural network (DNN) is used as a classifier model to allocate proper DR stages. An extensive set of experimental analyses takes place using a benchmark MESSIDOR dataset and the results are examined interms of different evaluation parameters. The simulation values highlighted the better DR diagnostic efficiency of the DL-DRDC technique over the recent techniques.