Apply a function per matrix cell for multiple matrices numpy

I’ll try to simplify my actual problem.
Let’s say I have 3 matrices with similar shape:

m1
array([[0.1 , 0.12, 0.32],
       [0.22, 0.15, 0.16],
       [0.42, 0.52, 0.62],
       [0.44, 0.35, 0.36]])

m2
array([[0.12 , 0.144, 0.384],
       [0.264, 0.18 , 0.192],
       [0.504, 0.624, 0.744],
       [0.528, 0.42 , 0.432]])

m3
array([[0.106 , 0.1272, 0.3392],
       [0.2332, 0.159 , 0.1696],
       [0.4452, 0.5512, 0.6572],
       [0.4664, 0.371 , 0.3816]])

Each matrix has a corresponded date, which I convert to an integer and call it delta, so:

delta_m1  = 2
delta_m2 = 5
delta_m3 = 10

What I would like to do is to apply a function to each cell-based-time-series, meaning that if I consider all cells at position (0,0), then m1_cell is 0.1, m2_cell is 0.12, and m3_cell is 0.106. So now I have a cell-based-time-series where the time (the x-axis) being the deltas (2,5,10) and the y-axis is the value of the cells (0.1,0.12,0.106). I would like to apply a smooth function (that takes two 1d-arrays and outputs 1d-array) that takes this cell-based-time-series, meaning takes X=(2,5,10) and y=(0.1,0.12,0.106) and outputs a new value per date/delta/matrix, so, for example, an output could be (0.09,0.1,0.8), where each value corresponds to the cell at (0,0) for each matrix. The whole process I’ve described refers to only one cell. I would like to apply it to all cells. Then I can create new matrices based on the results. I guess I can simply iterate over each cell over each matrix, and apply the function, but that seems super not efficient.
How can I achieve that in a pythonic efficient way?

EDIT

The function I would like to apply on each cell-based-time-series:

def local_regression(x0, X, Y, k):
    # add bias term
    x0 = np.r_[1, x0]
    X = np.c_[np.ones(len(X)), X]
    
    # fit model: normal equations with kernel
    xw = X.T * radial_kernel(x0, X, k)
    beta = np.linalg.pinv(xw @ X) @ xw @ Y
    
    # predict value
    return x0 @ beta

def radial_kernel(x0, X, k):
    return np.exp(np.sum((X - x0) ** 2, axis=1) / (-2 * k * k))

def smooth(x,y):
    return [local_regression(x[p],x,y,k) for p in range(len(x))]

So for example:

y = np.array([2,5,10])
x = np.array([0.1,0.12,0.106])
smooth(x,y)
>>> array([5.14556944, 6.34810085, 5.50632924])

Where x is an array of integers (though it can be changed to float) and y is an array of float. k is a parameter of the function and here k=20