Respiratory diseases remain a major cause of mortality worldwide, often resulting in reduced lung function and symptoms such as shortness of breath, coughing, and low blood oxygen levels. Early detection of these conditions is critical for timely treatment. There are various radiographic techniques, with X-ray imaging being a commonly used, cost-effective method that provides coverage of wide lung area. X-ray images often contain irrelevant regions that do not contribute to lung disease classification. Thus, segmenting the lung area is crucial to extract meaningful features for diagnosis. In this study, we use X-ray images as input and applied three customized segmentation models which are DeepLabV3, U-Net and FCNN (Fully Convolutional Neural Networks). Traditional segmentation models have complex structures which require more computational resources. In contrast, the proposed models are lightweight interms of number of layers and optimized for efficiency. The models are trained with different learning rates (0.0001 and 0.00001) and epochs (20 and 50) to evaluate their performance. Among them, U-Net with a learning rate of 0.0001 trained for 50 epochs, achieved the highest accuracy of 99.34%, IoU (Intersection over Union) of 98.6%, and Dice Coefficient of 99%. These results demonstrate that the proposed models effectively perform lung segmentation with fewer layers while maintaining high accuracy.