The novel coronavirus (SARS-CoV-2) has led to a pandemic. Due to its highly contagious nature, it has spread rapidly, resulting in major disruption to public health. In addition, it has also had a severe negative impact on the world economy. As a result, it is widely recognized now that widespread testing is key to containing the spread of the disease and opening up the economy. However, the current testing regime has been unable to keep up with testing demands. Hence, there is a need for an alternative approach for repeated large-scale testing of COVID-19. The emergence of wearable medical sensors (WMSs) and novel machine learning methods, such as deep neural networks (DNNs), points to a promising approach to address this challenge. WMSs enable continuous and user-transparent monitoring of the physiological signals. However, disease detection based on WMSs/DNNs and their deployment on resource-constrained edge devices remain challenging problems. In this work, we propose CovidDeep, a framework that combines efficient DNNs with commercially available WMSs for pervasive testing of the coronavirus. We collected data from 87 individuals, spanning four cohorts including healthy, asymptomatic (but SARS-CoV-2-positive) as well as moderately and severely symptomatic COVID-19 patients. We trained DNNs on various subsets of the features extracted from six WMS and questionnaire categories to perform ablation studies to determine which subsets are most efficacious in terms of test accuracy for a four-way classification. The highest test accuracy obtained was 99.4%. Since different WMS subsets may be more accessible (in terms of cost, availability, etc.) to different sets of people, we hope these DNN models will provide users with ample flexibility. The resultant DNNs can be easily deployed on edge devices, e.g., smartwatch or smartphone, which also has the benefit of preserving patient privacy.