The infection of respiratory coronavirus disease 2019 (COVID-19) starts with the upper respiratory tract and as the virus grows, the infection can progress to lungs and develop pneumonia. The conventional way of COVID-19 diagnosis is reverse transcription polymerase chain reaction (RT-PCR), which is less sensitive during early stages; especially if the patient is asymptomatic, which may further cause more severe pneumonia. In this context, several deep learning models have been proposed to identify pulmonary infections using publicly available chest X-ray (CXR) image datasets for early diagnosis, better treatment and quick cure. In these datasets, presence of less number of COVID-19 positive samples compared to other classes (normal, pneumonia and Tuberculosis) raises the challenge for unbiased learning of deep learning models. All deep learning models opted class balancing techniques to solve this issue; which however should be avoided in any medical diagnosis process. Moreover, the deep learning models are also data hungry and need massive computation resources. Therefore for quicker diagnosis, this research proposes a novel pinball loss function based one-class support vector machine (PB-OCSVM), that can work in presence of limited COVID-19 positive CXR samples with objectives to maximize the learning efficiency and to minimize the false predictions. The performance of the proposed model is compared with conventional OCSVM and existing deep learning models, and the experimental results prove that the proposed model outperformed over state-of-the-art methods. To validate the robustness of the proposed model, experiments are also performed with noisy CXR images and UCI benchmark datasets.