It is possible to make a mathematical model for the elbow point, at least if you have a smooth decreasing curve. In fact, here are three definitions of the elbow point, where we draw a line segment $A$ that connects the endpoints of the curve:
(1) Find the tangent line to the curve that is parallel to the line segment $A.$ Define the elbow point as the point where the tangent line intersects the curve.
(2) Find the point on the curve that has the greatest vertical distance to the line segment $A.$ Define the point as the elbow point.
(3) Find the point on the curve that has the greatest perpendicular distance to $A.$ Define this point as the elbow point.
Interestingly, under some surprisingly general conditions these definitions will all determine the same point.
Update: I eyeballed the OP's data and fit a power curve of the form $y=ax^{-b},$ which gives a great fit $(R^2 > 0.99)$. Using method (1) as described above, the elbow point is around 7.7 (so the user would likely select $k=8.$ Here is a visual: 
I also fit a logarithmic curve, (not shown here, but again $R^2 > 0.99$), which gave an elbow point of 7.99, in good agreement with the first fit.