I think that for this situation you can just represent your maze as a matrix of boolean values. The boolean values can be used to indicate whether there is a wall for that position.
Here's an example (with true indicating that the right side of the cell doesn't have a wall):
__ __ __ __
|__ |__ | [[true, false, true, false]
| | | | [false, false, true, false],
| | __| | => [false, true, false, false],
| | __| [true, false, true, false],
|__|__|__ __| [false, false, true, false]]
The given example matrix includes values for vertical walls. I only need to include information about the right direction because I can look at the cell at (x - 1) to get wall information for the left side of the cell.
You'll need a separate matrix for horizontal walls (for a total of two matrices). The top, bottom, and left edges implicitly have a wall.
Using this representation, you should be able to calculate your end point in a time proportional to the distance between an end and start point. The ball only needs to know its (x, y) position.
Your walls likely don't have to be their own class for this problem. If you ever want to extend your program such that landing on specific types of walls (say a fire wall) results in different behavior, then objects may be helpful.
Even in that case, I would say that using an enum may be sufficient.