During the 2nd surge of COVID-19 in India, the Delta variant caused a significant rise in infections and fatalities. Accurate diagnosis and timely treatment are crucial to control the spread and save lives. Common diagnostic tools include the Antigen Rapid Diagnostic Test (Ag-RDT), Reverse Transcription Polymerase Chain Reaction (RT-PCR) test, and CT scans. While Ag-RDT and RT-PCR can be performed by medical practitioners, their reliability is limited. As coronavirus disease mainly affects the respiratory system, chest X-rays can provide more accurate diagnosis. This study investigates the use of deep learning (DL) models in predicting coronavirus from chest X-ray images. Various pre-trained CNN models serve as feature vectors on both datasets, with the network heads updated for the current class labels of COVID or normal. By customizing and parameter tuning of pre-trained models, more than 90% prediction accuracy was achieved for VGG16, VGG19, ResNet50, and ResNet101. These models not only predict coronavirus disease cases from images but also determine the percentage likelihood of a given image being COVID-positive or normal. The implementation of Grad-CAM is utilized to highlight the class activation maps of images, which offers a more precise visualization of the affected areas of the lungs in coronavirus disease patients. The custom pre-trained models are remarkably efficient in accurately predicting coronavirus disease patients from both chest X-ray and radiography images. The methodology can be further improved to create an application that predicts the likelihood of a coronavirus disease case or normal case and produces visual explanations through activation maps.