An early effective screening and grading of COVID-19 has become imperative towards optimizing the limited available resources of the medical facilities. An automated segmentation of the infected volumes in lung CT is expected to significantly aid in the diagnosis and care of patients. However, an accurate demarcation of lesions remains problematic due to their irregular structure and location(s) within the lung. A novel deep learning architecture, Composite Deep network with Feature Weighting (CDNetFW), is proposed for efficient delineation of infected regions from lung CT images. Initially a coarser-segmentation is performed directly at shallower levels, thereby facilitating discovery of robust and discriminatory characteristics in the hidden layers. The novel feature weighting module helps prioritise relevant feature maps to be probed, along with those regions containing crucial information within these maps. This is followed by estimating the severity of the disease.The deep network CDNetFW has been shown to outperform several state-of-the-art architectures in the COVID-19 lesion segmentation task, as measured by experimental results on CT slices from publicly available datasets, especially when it comes to defining structures involving complex geometries.