This paper develops a customized residential load scheduling system to minimize electricity expenditure through optimizing the switching-on times of large energy-consuming appliances. Probability model of scheduled appliance is extracted from smart meter measurements using kernel density estimation method. Constrained switching-on time intervals as well as operation parameters of the appliance are then derived from the probability model. With time constraint and operation parameters, load scheduling is executed by solving a constraint optimization problem to minimize electricity expenditure under time-of-use tariff. The data-drive modeling strategy enables the load scheduling system to be adaptive to the diversity of residential customers. A customized time schedule thus can be produced in order to maximize economic benefits while meet the comfort requirement of residential customer.