A tropical cyclone (TC) making landfall at a populated coastal area can cause both loss of life and damage to infrastructure and the natural environment. Wind field models which accurately describe the spatial distribution of TC wind speeds can assist in reducing some of these risks, by identifying the most dangerous locations with respect to damaging winds. Existing models either have limitations in describing the skewness of the wind field or do not have the flexibility to incorporate multiple wind observations. We developed a new mathematical model to represent the temporal-spatial distribution of near-surface wind speeds in a TC over an open ocean that captures asymmetries in both the eyewall and tail of the wind profile. The model formulation is flexible, with options for a generic fit or inclusion of extra observations to achieve a fit with higher accuracy and has tuning parameters for calibration to wind observations. Utilising several covariates that could be obtained from satellite remote sensing data, we derive anchor points that the wind profile should pass through. Next, polar coordinate based cubic splines that minimise modified discontinuity energy and preserve specified shape constraints are used to model the wind speeds profile, which is formulated as solving several linear programming problems. Appropriate formulation adjustments are made for clockwise and counterclockwise rotations of air around a TC centre in the Southern and Northern Hemispheres, respectively. Model performance of the generic fit is tested using near-surface wind observations at automatic weather stations. Comparison with an earlier developed wind model shows that the spline model produces more accurate wind estimates near the radius of maximum winds. Although the spline model describes TC wind speeds over water well, a limitation is its inability to capture reduction in wind speeds caused by terrain friction. Overall, the spline model can be used successfully to describe TC wind speeds over waters.