Since the emergence of COVID-19 in December 2019, it has continued to ravage our world up until now. Many lives are at risk because of the menace caused by it. Although manual detection with the reverse transcription polymerase chain reaction (RT-PCR) is still popular and operational, it is costly and time-consuming. In this paper, we propose a novel transfer learning model using pre-trained InceptionV3 to classify infected people from healthy people using chest X-ray images. Even in a seemingly post-COVID era, the study of effective deep models for COVID detection and classification remains imperative, as one, the pandemic is likely to recur, two, monitor a few extant cases, and could be repurposed to other respiratory-based conditions. Methodically, by way of initial preprocessing, images were filtered using the non-local means (NLM) filter and improved in contrast using contrast-limited adaptive histogram equalization (CLAHE). Afterwards, data augmentation techniques with random affine transformation were adopted to augment the training data. Three experiments were undertaken on a publicly available dataset from Kaggle and in the end, our proposed model (modifiedInceptionV3) achieved a remarkable, state-of-the-art-accuracy of 98.6%.